A MULTI-VARIABLE THETA PRODUCT

BRUCE C. BERNDT1 AND ALEXANDRU ZAHARESCU

Abstract. A multi-variable theta product is examined. It is shown that, under very general choices
of the parameters, the quotient of two such general theta products is a root of unity. Special cases
are explicitly determined. The second main theorem yields an explicit evaluation of a sum of series of
cosines, which greatly generalizes one of Ramanujan’s theorems on certain sums of hyperbolic cosines.

1. Introduction
Motivated by a fascinating identity involving a quasi-theta product in Ramanujan’s
lost notebook [8], [4], we examine in this paper a certain, very general multi-variable
theta product. For k + 1 complex variables u1 , . . . , uk , w, k ≥ 1, define
Y∞
1 + (−1)j1 +···+jk wu2j
1
1 +1
· · · u2j
k
k +1

Fk (u1 , . . . , uk ; w) := . (1.1)
1 − (−1)j1 +···+jk wu2j1 +1 · · · u2jk +1
j1 ,...,jk =0 1 k

Throughout the paper, we assume that 0 < |u1 |, . . . , |uk | < 1. The product on the right
side of (1.1) converges for any w not of the form (−1)j1 +···+jk u1−2j1 −1 · · · u−2j k
k −1
with
j1 , . . . , jn ∈ N, and so, as a function of w, Fk (u1 , . . . , uk ; w) is meromorphic.
When k = 1, u1 = q, and w = q, then
Y∞ ∞ ∞
1 + (−1)j q 2j+2 Y 1 + q 4j+2 Y 1 − q 4j
F1 (q; q) = =
j=0
1 − (−1)j q 2j+2 j=0 1 − q 4j+2 j=1 1 + q 4j
Y∞
1 − q 8j−4 (1 − q 4j )2 ψ 2 (q 2 )
= = ,
j=1
1 − q 8j (1 − q 4j−2 )2 ψ(q 4 )
where we have used the well-known product representation [3, p. 36, Chap. 16, Entry
22(ii)] for the theta function ψ(q) (in Ramanujan’s notation),
X∞
ψ(q) := q n(n+1)/2 , |q| < 1.
n=0

Thus, in this instance, (1.1) reduces to a quotient of classical theta functions.

Section 2 is devoted to proving our two main results. Our first theorem shows that
for very general choices of the parameters 0 < |u1 |, . . . , |uk | < 1, the quotient of two
functions (1.1) is a root of unity when we take the two choices of w to be reciprocals
1
Research partially supported by grant MDA904-00-1-0015 from the National Security Agency.
2
Mathematics Subject Classification 2000: Primary, 11F27; Secondary, 33D10.
1
2 BRUCE C. BERNDT AND ALEXANDRU ZAHARESCU

of each other. Our second theorem gives an evaluation of a sum of k + 1 infinite series
involving cosines.
In Section 3, we provide applications. First, we show that Theorem 2.2 yields a vast
generalization of the following identity of Ramanujan found in Entry 15 of Chapter 14
of his second notebook [7], [2, p. 262]. Let α, β > 0 with αβ = π 2 /4. Then
∞
X X ∞
(−1)n (−1)n π
+ = . (1.2)
n=0
(2n + 1) cosh{(2n + 1)α} n=0 (2n + 1) cosh{(2n + 1)β} 4

We next examine special instances of Theorem 2.1. In particular, we give new explicit
evaluations of certain infinite products.

2. Multi-variable products
Our first objective is to prove the following transformation formula.
Theorem 2.1. Let A1 , . . . , Ak+1 and A be complex numbers such that A1 , . . . , Ak+1
are nonzero and the quotient of any two³ of them ´ is nonreal. For each distinct pair
πiAl
(l, j), 1 ≤ l, j ≤ k + 1, denote ql,j = exp ± 2Aj , where the sign ± is chosen so that
³ ´
πA
|ql,j | < 1. We also set qj = i exp 2A j
for 1 ≤ j ≤ k + 1. Then

k+1
Y µ ¶
Fk (q1,j , . . . , qj−1,j , qj+1,j , . . . , qk+1,j ; qj ) πi
= exp . (2.1)
F (q , . . . , qj−1,j , qj+1,j , . . . , qk+1,j ; q1j )
j=1 k 1,j
2k

The form of (2.1) is not surprising, for in the “inversion” formulas for classical theta
functions, roots of unity arise. We will derive Theorem 2.1 from the following result.
Theorem 2.2. Let A1 , . . . , Ak+1 be nonzero complex numbers with the quotient of any
two of them nonreal. Then for any complex number A, with |A| small enough in terms
of A1 , . . . , Ak+1 ,
(2n+1)πA
k+1 X
X ∞
(−1)n e 2Aj
π
Q (2n+1)πAl
= . (2.2)
j=1 n=−∞ (2n + 1) 1≤l≤k+1,l6=j cos
2
2Aj

Proof of Theorem 2.1. We first derive a convenient expression for log Fk (u1 , . . . , uk ; w).
Taking logarithms on both sides of (1.1), we find that
∞
X
log Fk (u1 , . . . , uk ; w) = log(1 + (−1)j1 +···+jk wu2j
1
1 +1
· · · u2j
k
k +1
) (2.3)
j1 ,...,jk =0
X ∞
− log(1 − (−1)j1 +···+jk wu2j
1
1 +1
· · · u2j
k
k +1
) =: S1 − S2 .
j1 ,...,jk =0

Here and in the next step, we have ignored branches of the logarithm. The justification
lies in our eventual proof of Theorem 2.2. Using the Taylor series of log(1 + z) about
 A MULTI-VARIABLE THETA PRODUCT 3

z = 0, we find that
∞
X X∞
((−1)1+j1 +···+jk wu2j
1
1 +1
· · · u2j
k
k +1 m
)
S1 = −
j1 ,...,jk =0 m=1
m
X∞ ∞
X
(−wu1 · · · uk )m
=− (−1)m(j1 +···+jk ) u2mj
1
1
· · · u2mj
k
k

m=1
m j1 ,...,jk =0
∞
à ∞
! Ã ∞
!
X (−wu1 · · · uk )m X X
=− (−u21 )mj1 ··· (−u2k )mjk
m=1
m j1 =0 jk =0
∞
X m
(−wu1 · · · uk )
=− . (2.4)
m=1
m(1 − (−u21 )m ) · · · (1 − (−u2k )m )
By replacing w by −w in (2.4), we find that
∞
X (wu1 · · · uk )m
S2 = − . (2.5)
m=1
m(1 − (−u21 )m ) · · · (1 − (−u2k )m )
Thus, substituting (2.4) and (2.5) into (2.3), we deduce that
∞
X (wu1 · · · uk )m (1 − (−1)m )
log Fk (u1 , . . . , uk ; w) =
m=1
m(1 − (−u21 )m ) · · · (1 − (−u2k )m )
∞
X (wu1 · · · uk )m
=2 . (2.6)
m=1
m(1 + u2m 2m
1 ) · · · (1 + uk )
m odd

Since u1 , . . . , uk are nonzero, we may write (2.6) in the form

∞
X w2n+1
log Fk (u1 , . . . , uk ; w) = 2 . (2.7)
n=0
(2n + 1)(u2n+1
1 + u1−2n−1 ) · · · (u2n+1
k + uk−2n−1 )
If we apply (2.7) with w replaced by 1/w and use the symmetry of the denominator
on the right side of (2.7) with respect to the transformation 2n + 1 → −(2n + 1) we
find that
µ ¶ −∞
X
1 w2n+1
log Fk u1 , . . . , uk ; = −2 2n+1 −2n−1 2n+1 −2n−1 .
w n=−1
(2n + 1)(u 1 + u 1 ) · · · (u k + u k )
(2.8)
Note that the series on the right side of (2.7) converges absolutely for any w with
|w| < 1/|u1 · · · uk |, while the series on the right side of (2.8) converges for any w with
|w| > |u1 · · · uk |. It follows that
µ ¶
1
log Fk (u1 , . . . , uk ; w) − log Fk u1 , . . . , uk ;
w
X∞
w2n+1
=2 , (2.9)
n=−∞
(2n + 1)(u2n+1 1 + u−2n−1
1 ) · · · (u2n+1
k + uk−2n−1 )
4 BRUCE C. BERNDT AND ALEXANDRU ZAHARESCU

for any w in the annulus

1
|u1 · · · uk | < |w| < . (2.10)
|u1 · · · uk |
Now recall that A1 , . . . , Ak+1 are nonzero complex numbers with the quotient of any
two of them nonreal. Also, recall that A is a complex number with |A| small enough
in terms of A1 , . . . , Ak+1 , such that (2.2) is valid and such that for any 1 ≤ j ≤ k + 1,
(2.10) holds with u1 , . . . , uk and w replaced by q1,j , . . . , q³
j−1,j , qj+1,j
´, . . . , qk+1,j , and
(2n+1)πA
qj , respectively. Then, taking into account that (−1)n exp 2Aj
= −iqj2n+1 and
2 cos (2n+1)πA
2Aj
l 2n+1
= ql,j −2n−1
+ ql,j , from (2.2) and (2.9), we deduce that

k+1 ³
X
log Fk (q1,j , . . . , qj−1,j , qj+1,j , . . . , qk+1,j ; qj )
j=1
³ 1 ´´ πi
− log Fk q1,j , . . . , qj−1,j , qj+1,j , . . . , qk+1,j ; = k. (2.11)
qj 2
By exponentiating both sides of (2.11), we obtain (2.1) for any A with |A| small
enough in terms of A1 , . . . , Ak+1 . Then the general form of (2.1) follows by analytic
continuation, and this completes the proof of Theorem 2.1. ¤

Proof of Theorem 2.2. Let A1 , . . . , Ak+1 and A be as in the statement of Theorem 2.2
and define the function
eAz
f (z) = .
z cos A1 z · · · cos Ak+1 z
The function f (z) is meromorphic in the entire complex plane with a simple pole at
z = 0 and simple poles at z = (2n + 1)π/(2Aj ) for each integer n and each integer
j, 1 ≤ j ≤ k + 1. Let γRm be a sequence of positively oriented circles centered at the
origin and with radii Rm tending to ∞ as m → ∞, where the radii Rm are chosen
so that the circles remain at a bounded distance from all the poles of f (z). From the
definition of f , it is easy to see that f (z) decays exponentially as |z| → ∞ provided z
remains at a bounded distance from the poles of f and |A| is small enough in terms of
A1 , . . . , Ak+1 . It follows that
¯Z ¯
¯ ¯
¯ ¯
¯ f (z)dz ¯ → 0, (2.12)
¯ γRm ¯

as Rm → ∞.
Let R(a) denote the residue of f (z) at a pole a. Then, brief calculations show that

R(0) =1, (2.13)

µ ¶ (2n+1)πA
(2n + 1)π 2(−1)n+1 e 2Aj
R = Q , (2.14)
2Aj π(2n + 1) 1≤l≤k+1,l6=j cos (2n+1)πA
2Aj
l
 A MULTI-VARIABLE THETA PRODUCT 5

for each integer n and each j ∈ {1, . . . , k + 1}. Hence, using (2.13), (2.14), and the
residue theorem, we deduce that
Z k+1
X X
(2n+1)πA
1 2(−1)n+1 e 2Aj
f (z)dz = 1 + Q (2n+1)πAl
. (2.15)
2πi γRm j=1 |2n+1|<2Rm Aj /π π(2n + 1) 1≤l≤k+1,l6=j cos 2Aj

Letting Rm tend to ∞ in (2.15) and employing (2.12), we conclude that

(2n+1)πA
k+1 X
X ∞
2(−1)n+1 e 2Aj
0=1+ Q (2n+1)πAl
. (2.16)
j=1 n=−∞ π(2n + 1) 1≤l≤k+1,l6=j cos 2Aj

This gives (2.2), and the theorem is proved. ¤

3. Applications
In this section we derive some applications of Theorems 2.1 and 2.2. We begin with
Theorem 2.2.
If we set k = 1 in Theorem 2.2, then, subject to the prescribed conditions on A, A1 ,
and A2 , (2.2) takes the shape
(2n+1)πA (2n+1)πA
∞
X ∞
X
(−1)n e 2A1 (−1)n e 2A2 π
(2n+1)πA2
+ (2n+1)πA1
= . (3.1)
n=−∞ (2n + 1) cos n=−∞ (2n + 1) cos 2
2A1 2A2

Set
πA2 πA1
A = 0, = iα (α > 0), and = −iβ (β > 0).
2A1 2A2
Thus, αβ = π 2 /4, and (3.1) reduces to the identity
∞
X X∞
(−1)n (−1)n π
+ = , (3.2)
n=−∞
(2n + 1) cosh{(2n + 1)α} n=−∞ (2n + 1) cosh{(2n + 1)β} 2

which is obviously equivalent to (1.2). As mentioned in the Introduction, this identity

can be found in Ramanujan’s second notebook [7, Chap. 14, Entry 15], [2, p. 262].
The first proof was given in 1925 by S. L. Malurkar [5], but because the notebooks
were not made available to the general public until 1957, he did not realize that he
had proven a result from Ramanujan’s notebooks. In 1951, T. S. Nanjundiah [6] gave
another proof. Finally, Berndt [1, pp. 176–177, Prop. 4.5] offered a further proof in
1977. We emphasize that all the authors cited above, in fact, found generalizations of
(3.2). The generalizations proved in this paper are different and seem to be new.
For a second example, let k = 2 and A = 0. For ω := exp(2πi/3), set
A2 A1 A3
= αω, = βω, and = γω,
A1 A3 A2
6 BRUCE C. BERNDT AND ALEXANDRU ZAHARESCU

where α, β and γ are positive real numbers. Thus, by Theorem 2.2, if α, β, and γ are
positive real numbers such that αβγ = 1, then
X∞
(−1)n
³ ´ ³ ´
(2n+1)παω (2n+1)πω 2
n=−∞ (2n + 1) cos 2
cos 2β
∞
X (−1)n
+ ³ ´ ³ ´
(2n+1)πβω (2n+1)πω 2
n=−∞ (2n + 1) cos 2
cos 2γ
∞
X (−1)n π
+ ³ ´ ³ ´= . (3.3)
n=−∞ (2n + 1) cos (2n+1)πγω
cos (2n+1)πω 2 2
2 2α

In particular, if α = β = γ = 1, then (3.3) reduces to

X∞
(−1)n π
³ ´ ³ ´= .
n=0 (2n + 1) cos
(2n+1)πω
cos (2n+1)πω
2
12
2 2

The latter two equalities and all such multi-sum identities arising from Theorem 2.2
are apparently new.
We now consider applications of Theorem 2.1.
We see from the definition of Fk that
1
Fk (u1 , . . . , uk ; −w) = . (3.4)
Fk (u1 , . . . , uk ; w)
If we set A = 0 in (2.1), then qj = i for 1 ≤ j ≤ k + 1. Using also (3.4), we find that
Ãk+1 !2 µ ¶
Y πi
Fk (q1,j , . . . , qj−1,j , qj+1,j , . . . , qk+1,j ; i) = exp k . (3.5)
j=1
2
Taking square roots on both sides of (3.5), we obtain
k+1
Y µ ¶
πi
Fk (q1,j , . . . , qj−1,j , qj+1,j , . . . , qk+1,j ; i) = ± exp . (3.6)
j=1
2k+1
In order to determine the sign ± on the right side of (3.6), we use (3.4) with w = i to
deduce that
log Fk (u1 , . . . , uk ; 1/i) = log Fk (u1 , . . . , uk ; −i) = − log Fk (u1 , . . . , uk ; i). (3.7)
Then, combining (2.9) and (3.7), we obtain
X∞
(−1)n
log Fk (u1 , . . . , uk ; i) = i 2n+1 −2n−1 2n+1 −2n−1 . (3.8)
n=−∞
(2n + 1)(u 1 + u 1 ) · · · (u k + u k )
In particular, for any j ∈ {1, . . . , k + 1}, the equality (3.8) gives
∞
i X (−1)n
log Fk (q1,j , . . . , qj−1,j , qj+1,j , . . . , qk+1,j ; i) = k Q .
2 n=−∞ (2n + 1) 1≤l≤k+1,l6=j cos (2n+1)πAl
2Aj
(3.9)
 A MULTI-VARIABLE THETA PRODUCT 7

If we add the equalities (3.9) for j = 1, . . . , k + 1, then exponentiate both sides and use
(2.2), we obtain
k+1
Y µ ¶
πi
Fk (q1,j , . . . , qj−1,j , qj+1,j , . . . , qk+1,j ; i) = exp k+1 . (3.10)
j=1
2
¡ ¢
An interesting particular case arises when Aj = exp 2πij k+1
, 1 ≤ j ≤ k + 1. We
need the quotient of any two of them to be nonreal, and for this reason we choose k
to be even, say k = 2r. Note that for any j ∈ {1, . . . , 2r + 1} the numbers A1 /Aj ,
A2 /Aj , . . . , Aj−1 /Aj , Aj+1 /Aj , . . . , A2r+1 /Aj coincide in a certain order with the num-
bers A1 , . . . , A2r . Therefore, the set {q1,j , . . . , qj−1,j , qj+1,j , . . . , q2r+1,j } is the same for
any j and coincides with the set {exp (±πiA1 /2) , . . . , exp (±πiA2r /2)}, where the signs
2πj
are chosen such that | exp (±πiAj /2) | < 1, 1 ≤ j ≤ 2r. Since Im Aj = sin 2r+1 is posi-
tive for 1 ≤ j ≤ r and is negative for r + 1 ≤ j ≤ 2r, the choice of the signs above will
be “plus” for 1 ≤ j ≤ r and “minus” for r + 1 ≤ j ≤ 2r. From the definition (1.1) we
see that Fk (u1 , . . . , uk ; w) is symmetric in u1 , . . . , uk . In our particular case, it follows
that all the factors on the left side of (3.10) are equal, and coincide with
¡ ¢
F2r eπiA1 /2 , . . . , eπiAr /2 , e−πiAr+1 /2 , . . . , e−πiA2r /2 ; i .
Thus (3.10) gives
µ ¶
¡ ¡ πiA1 /2 πiAr /2 −πiAr+1 /2 −πiA2r /2
¢¢2r+1 πi
F2r e ,...,e ,e ,...,e ;i = exp . (3.11)
22r+1
Then from (3.11) it follows that there is an integer h, 0 ≤ h ≤ 2r, such that
µ ¶
¡ πiA1 /2 πiAr /2 −πiAr+1 /2 −πiA2r /2
¢ πi 2hπi
F2r e ,...,e ,e ,...,e ; i = exp + .
(2r + 1)22r+1 2r + 1
(3.12)
In order to determine the value of h in (3.12), we again rely on Theorem 2.2. Since in
our case the left side of (3.9) is the same for any j ∈ {1, . . . , 2r + 1} and coincides with
¡ ¢
log F2r eπiA1 /2 , . . . , eπiAr /2 , e−πiAr+1 /2 , . . . , e−πiA2r /2 ; i ,
by combining (3.9) and (2.2) we find that
¡ ¢ πi
log F2r eπiA1 /2 , . . . , eπiAr /2 , e−πiAr+1 /2 , . . . , e−πiA2r /2 ; i = . (3.13)
(2r + 1)22r+1
Exponentiating both sides of (3.13), we find that
µ ¶
¡ πiA1 /2 πiAr /2 −πiAr+1 /2 −πiA2r /2
¢ πi
F2r e ,...,e ,e ,...,e ; i = exp . (3.14)
(2r + 1)22r+1
Note that the left side of (3.14) has the form
¡ ¢
F2r ρ1 eiθ1 , ρ2 eiθ2 , . . . , ρr eiθr , ρr e−iθr , . . . , ρ2 e−iθ2 , ρ1 e−iθ1 ; i ,
2πj 2πj
where ρj = exp(− π2 sin 2r+1 ) and θj = π2 cos 2r+1 , for 1 ≤ j ≤ r. ¿From the definition
(1.1) we see that for any real numbers ρ1 , . . . , ρr , θ1 , . . . , θr with 0 < ρj < 1 for 1 ≤ j ≤
8 BRUCE C. BERNDT AND ALEXANDRU ZAHARESCU

r,
¡ ¢
F2r ρ1 eiθ1 , ρ2 eiθ2 , . . . , ρr eiθr , ρr e−iθr , . . . , ρ2 e−iθ2 , ρ1 e−iθ1 ; i (3.15)
∞ Q ¡ ¢2jm +1 ¡ ¢2lm +1
Y 1 + i(−1)j1 +···+jr +l1 +···+lr rm=1 ρm eiθm ρm e−iθm
= Q
j1 ,l1 ,...,jr ,lr =0
1 − i(−1)j1 +···+jr +l1 +···+lr rm=1 (ρm eiθm )2jm +1 (ρm e−iθm )2lm +1
∞
Y 2(j1 +l1 +1) 2(jr +lr +1) 2iθ1 (j1 −l1 )
1 + i(−1)j1 +···+jr +l1 +···+lr ρ1 · · · ρr e · · · e2iθr (jr −lr )
= 2(j1 +l1 +1) 2(jr +lr +1) 2iθ1 (j1 −l1 )
.
j1 +···+jr +l1 +···+lr ρ · · · e2iθr (jr −lr )
j1 ,l1 ,...,jr ,lr =0 1 − i(−1) 1 · · · ρr e
In our case we may write
à r
!
2(j1 +l1 +1)
X 2πm
ρ1 · · · ρ2(j
r
r +lr +1)
= exp −π (jm + lm + 1) sin (3.16)
m=1
2r + 1
and à !
r
X 2πm
e2iθ1 (j1 −l1 ) · · · e2iθr (jr −lr ) = exp πi (jm − lm ) cos . (3.17)
m=1
2r + 1
Combining (3.14)–(3.17), we obtain the following result.
Theorem 3.1. For each integer r ≥ 1,
Pr Pr
∞
Y 2πm 2πm
1 + i(−1)j1 +···+jr +l1 +···+lr e−π m=1 (jm +lm +1) sin 2r+1 +πi m=1 (jm −lm ) cos 2r+1
Pr 2πm Pr 2πm
j1 +···+jr +l1 +···+lr e−π m=1 (jm +lm +1) sin 2r+1 +πi m=1 (jm −lm ) cos 2r+1
j1 ,l1 ,...,jr ,lr =0 1 − i(−1)
µ ¶
πi
= exp . (3.18)
(2r + 1)22r+1
Let us write down explicitly the left side of (3.18) for the first two values of r. If
r = 1, then (3.18) reduces to
Y∞
1 + i(−1)j+l exp (−π(j + l + 1) sin(2π/3) + πi(j − l) cos(2π/3))
j+l exp (−π(j + l + 1) sin(2π/3) + πi(j − l) cos(2π/3))
= eπi/24 . (3.19)
j,l=0
1 − i(−1)
On the left side of (3.19), make the substitution n = j + l + 1 and note that
µ ¶ µ ¶
2π πi(l − j)
exp πi(j − l) cos = exp = il−j = in−1−2j = in−1 (−1)j .
3 2
Then (3.19) becomes
√
∞ n−1
Y Y 1 + in (−1)n−1+j e−πn 3/2
√ = eπi/24 . (3.20)
n=1 j=0
1− in (−1)n−1+j e−πn 3/2
In the inner product on the left side of (3.20), the contributions of any two consecutive
values of j cancel each other. Since j takes exactly n values, we deduce that, if n is
even, then the inner product above equals 1, while if n is odd, say n = 2k + 1, then
the inner product equals
√ √
1 + in (−1)n−1 e−πn 3/2 1 + i(−1)k e−π(2k+1) 3/2
√ = √ .
1 − in (−1)n−1 e−πn 3/2 1 − i(−1)k e−π(2k+1) 3/2
 A MULTI-VARIABLE THETA PRODUCT 9

Thus (3.20) reduces to

√
∞
Y 1 + i(−1)k e−π(2k+1) 3/2
√ = eπi/24 . (3.21)
k=0
1− i(−1)k e−π(2k+1) 3/2
The evaluation (3.21) can be formulated in terms of Dedekind eta-functions. Recall
that, for q = exp(2πiτ ), Im τ > 0, the Dedekind eta-function η(τ ) is defined by
∞
Y
2πiτ /24
η(τ ) := e (1 − e2πinτ ).
n=1

A brief calculation shows that

∞ ¡1 ¢
Y
n+1/2 1/48 η 2
τ
(1 − q )=q . (3.22)
n=0
η(τ )
√
Observe that if we set τ = ± 12 + 12 i 3 in (3.22), we obtain, respectively, the denominator
and numerator in the product (3.21).
This last part of the reasoning works in more generality. To be precise, for any
complex number u with |u| < 1 and any w for which F2 (iu, u; w) is defined, we can use
the same reasoning as above to show that
∞
Y 1 + iwu4k+2
F2 (iu, u; w) = .
k=0
1 − iwu4k+2
Returning to (3.18), in the case r = 2, we obtain
∞ P2 P2
Y 1 + i(−1)j1 +j2 +l1 +l2 e−π m=1 (jm +lm +1) sin(2πm/5)+πi m=1 (jm −lm ) cos(2πm/5)
P P = eπi/160 .
1 − i(−1)j1 +j2 +l1 +l2 e−π 2m=1 (jm +lm +1) sin(2πm/5)+πi 2m=1 (jm −lm ) cos(2πm/5)
j1 ,j2 ,l1 ,l2 =0

√ √ (3.23)
√ √
10+2 5 10−2 5
We
√
now use the equalities sin(2π/5) √
= 4
, sin(4π/5) = 4
, cos(2π/5) =
5−1 −1− 5
4
, and cos(4π/5) = 4
, and make the substitutions n1 = j1 + l1 + 1, n2 =
j2 + l2 + 1, n = j1 − j2 , and j = j1 . Then the numerator on the left side of (3.23) equals
√ √ √ √ √
n1 +n2 − π4 (n1 10+2 5+n2 10−2 5)+ πi ((n2 −n1 +2j1 −2j2 ) 5+n1 +n2 −2j1 −2j2 −2)
1 + i(−1) e 4
√ √ √ √ √
−j1 −j2 n1 +n2 − π4 (n1 10+2 5+n2 10−2 5)+ πi ((n2 −n1 +2j1 −2j2 ) 5+n1 +n2 )
=1 + i (−1) e 4

π
√ √ √ √ πi
√
=1 + in (−1)n1 +n2 +j e− 4 (n1 10+2 5+n2 10−2 5)+ 4 ((n2 −n1 +2n) 5+n1 +n2 ) . (3.24)
If we denote
A = {(n1 , n2 , n, j) : j1 , j2 , l1 , l2 ∈ N}, (3.25)
then, by (3.24), (3.23) becomes
√ √ √ √ √
Y π
1 + in (−1)n1 +n2 +j e− 4 (n1 10+2 5+n2 10−2 5)+ πi
4
((n2 −n1 +2n) 5+n1 +n2 )

π
√ √ √ √ √ = eπi/160 .
10+2 5+n2 10−2 5)+ πi
(n1 ,n2 ,n,j)∈A 1− in (−1)n1 +n2 +j e− 4 (n1 4
((n2 −n1 +2n) 5+n1 +n2 )

(3.26)
10 BRUCE C. BERNDT AND ALEXANDRU ZAHARESCU

A brief calculation shows that

A = {(n1 , n2 , n, j) ∈ Z4 : n1 , n2 ≥ 1, 1 − n2 ≤ n ≤ n1 − 1,
max{n, 0} ≤ j ≤ min{n + n2 − 1, n1 − 1}}. (3.27)

Let us remark that for fixed n1 , n2 , n the contributions of any two consecutive values
of j on the left side of (3.26) cancel each other. If we denote

n∗ = min{n + n2 − 1, n1 − 1} − max{n, 0}, (3.28)

then j takes exactly n∗ + 1 values. We deduce that for any fixed n1 , n2 , and n as above,
the product over j on the left side of (3.26) equals 1 if n∗ is odd. If n∗ is even, then
the value of this product can be obtained by letting j equal any one of its two extreme
values, that is, j = max{n, 0} or j = min{n + n2 − 1, n1 − 1}. Thus, choosing the first
of these values for j and denoting for convenience

B = {(n1 , n2 , n, ) ∈ Z3 : n1 , n2 ≥ 1, 1 − n2 ≤ n ≤ n1 − 1, n∗ even}, (3.29)

we finally obtain
√ √ √ √ √
Y π
1 + in (−1)n1 +n2 +max{n,0} e− 4 (n1 10+2 5+n2 10−2 5)+ πi
4
((n2 −n1 +2n) 5+n1 +n2 )

π
√ √ √ √ √ = eπi/160 .
10+2 5+n2 10−2 5)+ πi
(n1 ,n2 ,n)∈B 1− in (−1)n1 +n2 +max{n,0} e− 4 (n1 4
((n2 −n1 +2n) 5+n1 +n2 )

(3.30)

4. Concluding remarks
The product on the left side of (3.21) involves one variable k, while the product on
the left side of (3.30) involves three variables n1 , n2 and n. For a general r, if one uses
the definition (1.1) on the left side of (3.14), this will involve products of the form
¡ πiA1 /2 ¢2j1 +1 ¡ ¢2jr +1 ¡ −πiAr+1 /2 ¢2jr+1 +1 ¡ ¢2j2r +1
e · · · eπiAr /2 e · · · e−πiA2r /2 := e−πiγ/2 (4.1)
2πi
say, where γ belongs to the ring of integers of the cyclotomic field Q(e 2r+1 ). The degree
2πi
of Q(e 2r+1 ) over Q equals ϕ(2r + 1), and each γ as above can be written uniquely in
the form
2πi 2πi(ϕ(2r+1)−1)
γ = a0 + a1 e 2r+1 + · · · + aϕ(2r+1)−1 e 2r+1

with a0 , a1 , . . . , aϕ(2r+1)−1 ∈ Z. Thus one may write the left side of (3.14) as a product
over the independent integral variables a0 , a1 , . . . , aϕ(2r+1)−1 , and a finite inner product,
over those (j1 , . . . , j2r ) ∈ N2r which correspond to a given (a0 , a1 , . . . , aϕ(2r+1)−1 ) ∈
Zϕ(2r+1) . Moreover a0 can be eliminated in the sense that its contribution to the
right side of (4.1) can be easily described reasoning mod 4. So we are left with
a1 , . . . , aϕ(2r+1)−1 . This explains why in our most simplified concrete equalities (3.21)
and (3.30) we had products over one and respectively three variables.
The authors are grateful to the referee for two very helpful suggestions.
 A MULTI-VARIABLE THETA PRODUCT 11

References
[1] B. C. Berndt, Modular transformations and generalizations of several formulae of Ramanujan,
Rocky Mt. J. Math. 7 (1977), 147–189.
[2] B. C. Berndt, Ramanujan’s Notebooks, Part II, Springer–Verlag, New York, 1989.
[3] B. C. Berndt, Ramanujan’s Notebooks, Part III, Springer–Verlag, New York, 1991.
[4] B. C. Berndt and A. Zaharescu, A quasi-theta product in Ramanujan’s lost notebook,
Math. Proc. Cambridge Philos. Soc., to appear.
[5] S. L. Malurkar, On the application of Herr Mellin’s integrals to some series, J. Indian
Math. Soc. 16 (1925–26), 130–138.
[6] T. S. Nanjundiah, Certain summations due to Ramanujan, and their generalizations, Proc. Indian
Acad. Sci., Sect. A 34 (1951), 215–228.
[7] S. Ramanujan, Notebooks (2 volumes), Tata Institute of Fundamental Research, Bombay, 1957.
[8] S. Ramanujan, The Lost Notebook and Other Unpublished Papers, Narosa, New Delhi, 1988.

Department of Mathematics, University of Illinois, 1409 West Green Street, Ur-

bana, IL 61801, USA
E-mail address: berndt@math.uiuc.edu

Department of Mathematics, University of Illinois, 1409 West Green Street, Ur-

bana, IL 61801, USA
E-mail address: zaharesc@math.uiuc.edu